基于SSA-ARIMA-BPNN组合模型的黄金价格预测

Gold Price Prediction Based on SSA-ARIMA-BPNN Combination Model

摘要: 考虑到金价预测具有极大的实用价值，建立了基于奇异谱分析(SSA)的黄金价格预测组合模型。由于黄金价格数据具有尖峰厚尾的特点，噪声含量高，首先使用奇异谱分析方法对数据进行分解与重构。然后分别对原始数据和重构数据建立单一ARIMA模型、单一BP神经网络模型，比较预测结果发现奇异谱分析可以消除数据的噪声，提高模型的预测精度。由于单一模型具有一定的局限性，建立了SSA-ARIMA(3,1,0)-BPNN组合模型进行黄金价格预测。实验结果表明，该组合模型有效提取了数据的信息，预测精度整体优于单一模型。

Abstract: Considering the great practical value of gold price prediction, a gold price prediction combination model based on singular spectrum analysis (SSA) was established. Due to the characteristic of sharp peaks and thick tails in gold price data, the noise content is high, singular spectrum analysis was first used to decompose and reconstruct the data. Then, a single ARIMA model and a single BP neural network model were established for the original and reconstructed data respectively. Comparing the prediction results, it was found that singular spectrum analysis can eliminate data noise and improve the prediction accuracy of the model. Due to the limitations of the single model, the SSA-ARIMA(3,1,0)-BPNN combination model is established to forecast the gold price. The experimental results show that the combined model effectively extracts the information of the data, and the overall prediction accuracy is better than the single model.